In a call with analysts during the company’s fiscal quarter 4 2025 earnings call Huang revealed that the new hardware would be called Blackwell Ultra and will be unveiled in March 2025 at the Nvidia GTC event.
Huang said, “Come to GTC, and I will talk to you about Blackwell Ultra and Vera Rubin and then show you one click after that.”
Nvidia Blackwell Ultra…and More
Huang added that Blackwell Ultra would be launched in the second quarter of 2025 and will offer upgrades in processors and networking. It will also be built on the Blackwell system architecture.
Both Blackwell Ultra, and Vera Rubin, were teased in a company roadmap at Computex 2020. Huang has now confirmed the names of both. Nvidia announced Blackwell at GTC 2020, promising a major leap forward in terms AI power and efficiency.
Since then, Nvidia has released a number releases, including the Blackwell “superchip”GB200 which can scale from a single rack to an entire datacenter.
Blackwell has 208 billion transistors on its two GPUs (up from the 80 billion in Hopper), which are linked by a 10 TB/second chip to chip link to create a single, unified, GPU. This makes it up to 30x quicker than Hopper for AI inference tasks and offers up to 20 petaflops in FP4 power. This is far ahead of any other product available today. Nvidia claims that Blackwell can reduce energy and cost consumption by 25x. For example, training a model with 1.8 trillion parameters, which would have previously required 8,000 Hopper graphics cards and 15 megawatts, can now be achieved by 2,000 Blackwell graphics cards using only four megawatts. Vera Rubin, the *next* step for Nvidia, is expected to be released in 2026. CPU and GPU products are anticipated, including a Vera Rubin Board combining the GPU with the CPU “superchip”.
Huang teased the announcement of Nvidia’s latest financial results. The company is once again experiencing record returns, thanks to the surge in AI demand.
Nvidia’s overall revenues have more than doubled in the past year, reaching $130.5 Billion for the financial year. This is an 114% increase, largely due to record data center revenues, which reached $35.6 Billion, up 16% on the previous quarter and 93% on a year ago.
The demand for Blackwell is incredible as reasoning AI adds a new scaling law – increasing compute for models makes them smarter, and increasing compute to do long thinking makes answers smarter,” Huang “The results are ,” said.
We’ve successfully ramped-up the massive-scale manufacturing of Blackwell AI Supercomputers, achieving millions of dollars in sales during its first quarter. AI is advancing at a rapid pace as agentic AI, physical AI and the next wave of AI revolutionizes the largest industries.
